ASUS R9 290X DirectCU II OC Overclocking Review @ [H]

KickAssCop said:
The 780 Ti rapes.
Click to expand...


Yes, it does. I'm very happy I sold my 290 CFX setup and went with a single 780ti. I haven't even overclocked it yet, it doesn't drop below 1020mhz even after 3 or 4 hours of BF4.
 
Yakk said:
Question :

Can the 290(x) chip possibly be paired up with faster memory? With the ultra-wide bus this could make for a drastic speed increase, I'm curious if there may be a technical reason (TDP?), or is it just cost? (At which point it opens up further possibilities for AIBs)
Click to expand...

It can be, but it's usually not the memory chips holding it back. AFAIK, all the Hawaii cards already come with chips rated for 1500MHz (6.0GHz effective). AMD simplified the memory controller to save silicon and that simplified memory controller is the limiting factor and the reason why many cards can't hit the memory speed their memory chips are rated for.

That is actually a really smart move by AMD since they are not bandwidth restricted as it is.
 
Nenu said:
. Go somewhere that is.
If you continue like this, I will report your posts.
Click to expand...

Seriously!!!....dude you seem a little too sensitive if my posts offend.....but of course that is a matter that is also very subjective.....very much like this article....after all there has been no measurement of the sound levels to verify that these cards are even running at similar sound levels....additionally who says what is an appropriate sound level?....some are more sensitive to this sort of thing than others, circumstances also vary, I wear headphone and don't care what the fan speed goes to as I can't hear it...there is also no measurement of the sound level that we can compare to other existing cards....so in all we have nothing but the maximum performance level at some arbitrary and unknown sound level and no adequate measure to say they are even operating at the same sound level....really!!.... this is so open to skewing the results in favour of what ever bias you have...and this site is notorious for it's Nvidia centric views.....now "IF" I was screaming "PAID FOR NVIDIA SHILLS" you'd have right to complain....but I'm not...I'm just politely pointing out this review is far too subjective and dependent on arbitrary restrictions to hold any weight.

So just to point out again.....when the cooling solution is sufficient and not hamstrung by arbitrary restrictions overclocked 290X has higher performance than overclocked 780ti....just get over it.
 
Last edited:
specops said:
It can be, but it's usually not the memory chips holding it back. AFAIK, all the Hawaii cards already come with chips rated for 1500MHz (6.0GHz effective). AMD simplified the memory controller to save silicon and that simplified memory controller is the limiting factor and the reason why many cards can't hit the memory speed their memory chips are rated for.

That is actually a really smart move by AMD since they are not bandwidth restricted as it is.
Click to expand...

I can clock my ram faster than 6GHz with +100mV on the core.
I benched it at 1620MHz, 5MHz short of the maximum allowed on MSI Afterburner.
Quad pumped gives 6.48GHz.
It ran fine at this speed but when returning to idle voltages while the memory was still hot, my machine became a bit unstable.

So the memory overclock has to also be stable at the voltage set when idle, not just the load voltage.
At least with the XFX 290X bios I am using, I have a feeling the Asus bios has some differences.
And overclocking with CCC must be disabled because on reboot it will apply the high memory overclock but cannot change to a supporting voltage. Instant machine crash when CCC loads.
I had this problem and my only resolution was to reboot in safe mode, disable CCC, reboot, un-install driver, reboot, re-install driver.
 
Nenu said:
I can clock my ram faster than 6GHz with +100mV on the core.
I benched it at 1620MHz, 5MHz short of the maximum allowed on MSI Afterburner.
Quad pumped gives 6.48GHz.
It ran fine at this speed but when returning to idle voltages while the memory was still hot, my machine became a bit unstable.

So the memory overclock has to also be stable at the voltage set when idle, not just the load voltage.
At least with the XFX 290X bios I am using, I have a feeling the Asus bios has some differences.
And overclocking with CCC must be disabled because on reboot it will apply the high memory overclock but cannot change to a supporting voltage. Instant machine crash when CCC loads.
I had this problem and my only resolution was to reboot in safe mode, disable CCC, reboot, un-install driver, reboot, re-install driver.
Click to expand...

The very fact that Hawaii memory OC seems to be closely connected with core voltage (this is a common trend, not just you) is evidence that the limiting factor is indeed on the GPU itself and not the memory chips. Sure, some Elpida chips might limit some cards that would be otherwise capable of going well over 1500MHz but on the whole, higher rated memory wouldn't do much good.
 
Yeah, it seems the memory controller is on the GPU die.
 
technogiant said:
So just to point out again.....when the cooling solution is sufficient and not hamstrung by arbitrary restrictions overclocked 290X has higher performance than overclocked 780ti....just get over it.
Click to expand...

Prove it. If this information is widely available like you seem to be implying, it shouldn't be hard.
 
LigTasm said:
Prove it. If this information is widely available like you seem to be implying, it shouldn't be hard.
Click to expand...

Just google around for watercooled benchmarks....I don't have any to hand atm....but that was the take home message I got from reading several different ones.
 
technogiant said:
I don't think it is fair testing these cards at different fan speeds....the only way to fairly test them is with both with 100% fans regardless of noise.

You say there was no throttling as you kept temps below 94c....that is debatable, power tune alters settings in fractions of a second and if you looked at this microscopically you may well find that throttling is occurring.

Secondly your selection of a lower fan speed for the AMD card may well be affecting the maximum overclock you are achieving....I would say this is undoubtedly happening as you are hitting 90c....higher silicon temps mean greater leakage which means more voltage is required to be stable which means more heat...you have basically surpassed the ability of the cooling solutions effectiveness.....its a vicious circle which you have deliberately brought about in the AMD card not only by using less than 100% fan but also using a slower fan speed than the Nvidia card you are comparing it to.....how can you ever justify that when making a "fair" comparison.
Click to expand...

90c degree was the peak temperature obtained, it wasn't the average.

Moving the fan to 100% did NOT change performance at 1115MHz. It added a whole lot of extra noise, with no performance benefits. Therefore, a waste.

You cannot compare percentage fan speeds between the ASUS card and the 780 Ti reference, completely different fans. Also completely different number of fans. That comparison is not valid.
 
EasterEEL said:
Interesting indeed.

If I am understanding correctly the stock cooled 780Ti is beating the custom cooled 290X by a fair margin in BF4 & C4 when both are overclocked and after the "warm up period" when the 780ti is throttling back to 941MHz.
Click to expand...

It drops to 941 at its stock speeds.

When we overclocked it to 1163 and pushed the fan to 80% it sustained 1163 clock speed for all the testing, no throttling down from 1163 with the fan that high.
 
technogiant said:
^^.....I see no reason or logic in comparing two cards when you use unequal settings.....if you are not going to use 100% fans on both...which you should in a performance test....this is a performance test not an acoustic test isn't it?....then at the very least use the same fan speeds on both cards, and no the author did not explain why he used a lower fan speed on the 290x than on the 780ti.....only why he didn't use 100% fan.

Running the 290x at 90c and the 780ti at 75c is ridiculous....the lower temp will obviously allow for a higher stable overclock....as I stated, whether you accept it or not, if adequate cooling is not an issue 290x outperforms 780ti
Click to expand...

You seem to be ignoring a lot of what I said in the article. I explained why the choices.

I also explained that a higher fan speed did not affect performance or allow a better overclock on the 290X. 100% fan yielded no benefits. But it did yield extremely loud noise that no one in their right mind would deal with.
 
Yakk said:
Question :

Can the 290(x) chip possibly be paired up with faster memory? With the ultra-wide bus this could make for a drastic speed increase, I'm curious if there may be a technical reason (TDP?), or is it just cost? (At which point it opens up further possibilities for AIBs)
Click to expand...

Well sure, but, I don't think Hawaii's limitations are its memory bandwidth. You have to balance engine and memory performance, just adding gobs of bandwidth won't necessarily net you a performance increase if the engine can't keep up.

It is analogous to having faster DDR on your motherboard, DDR3 1600 vs. 1866 vs. 2133, increasing that bandwidth does not yield a better gameplay experience, at most a frame or two better, that's it. Same with bandwidth on a GPU.
 
technogiant said:
Just google around for watercooled benchmarks....I don't have any to hand atm....but that was the take home message I got from reading several different ones.
Click to expand...

I see, we're talking about cherry picked BIOS modded 1250+ cards. Those are the norm now I suppose. Don't get me wrong here, I owned a pair of Hawaii cards under water. They were fast, but did not overclock well even with the Asus BIOS on them. They were average 1100-1125 chips like the vast majority of them are. 780ti has a ton of headroom, just like all of the GK110 chips, and if you want to get into modded BIOS cards look at what the zombie Titans can do. Those aren't representative of what everyone can buy though.
 
Brent_Justice said:
Moving the fan to 100% did NOT change performance at 1115MHz. It added a whole lot of extra noise, with no performance benefits. Therefore, a waste.
Click to expand...

That may be the case but you would have got a higher stable overclock at 100% fan speed.

The 290X having a smaller die size and so greater heat density is going to be very susceptible to reducing the cooling efficiency as you have done by limiting the fan speed.
Additionally running at higher temperature only increase the severity of that issue as increased silicon leakage at higher temps just leads to more power draw and more heat and less stability.

Don't miss understand me...even with these custom cooled solutions it may be that air cooling is simply insufficient to adequately cool this chip....but from what I've seen with water cooled benchmarks the overclocked 290X outperforms overclocked 780ti ie quite the converse to your findings here.

It would just have been nice to see a comparison of custom cooled 290X vs custom cooled 780ti at 100% fan just to see the all out winner.....as it stands this is just a comparison at some arbitrary and unmeasured sound level, circumstances that will most definitely favour the more easily cooled Nvidia chip.
 
Well Brent said that raising the fan to 100% didn't help so that's that. If the R9 290x was selling at MSRP then it still would be the value leader. But at the current inflated pricing you might as well get a 780 Ti if you're strictly gaming. The only advantages I see on the 290x are the OpenCL performance which is great for doing other tasks and the possibility of Mantle. But Mantle seems to be launching from a planet far, far, away...
 
technogiant said:
That may be the case but you would have got a higher stable overclock at 100% fan speed.

The 290X having a smaller die size and so greater heat density is going to be very susceptible to reducing the cooling efficiency as you have done by limiting the fan speed.
Additionally running at higher temperature only increase the severity of that issue as increased silicon leakage at higher temps just leads to more power draw and more heat and less stability.

Don't miss understand me...even with these custom cooled solutions it may be that air cooling is simply insufficient to adequately cool this chip....but from what I've seen with water cooled benchmarks the overclocked 290X outperforms overclocked 780ti ie quite the converse to your findings here.

It would just have been nice to see a comparison of custom cooled 290X vs custom cooled 780ti at 100% fan just to see the all out winner.....as it stands this is just a comparison at some arbitrary and unmeasured sound level, circumstances that will most definitely favour the more easily cooled Nvidia chip.
Click to expand...

Well this isn't an article about how well these cards could behave when water cooled. I doubt that you're going to see a review of water cooled cards here as just not enough people do it.
 
$699 at newegg... jesus christ. Will newegg include a cuddle after ass raping you with the 150 mark up?
 
cageymaru said:
Well this isn't an article about how well these cards could behave when water cooled. I doubt that you're going to see a review of water cooled cards here as just not enough people do it.
Click to expand...

I know this....I only refer to watercooling as the results are at difference with what has been stated here.

As I've conceded it may be that using less efficient air cooled solutions may reverse this result as the 290x is more difficult to cool....but this review has not proven that...and has in-fact further imposed artificial restrictions on the cooling solutions that would tend to favour the more easily cooled Nvidia chip/card.
 
technogiant said:
That may be the case but you would have got a higher stable overclock at 100% fan speed.
Click to expand...

Tried it. Apparently you did not read page 3.

I also tried 100% fan at 1.35v. End result, it did not.

It would just have been nice to see a comparison of custom cooled 290X vs custom cooled 780ti
Click to expand...

We have a custom cooled 780 Ti review in the works. It will be overclocked vs. the ASUS custom cooled 290X.

as it stands this is just a comparison at some arbitrary and unmeasured sound level, circumstances that will most definitely favour the more easily cooled Nvidia chip.
Click to expand...

It is not arbitrary. Nor does it favor one or the other. The ASUS card is not being held back in its overclock or its performance. This is about the 4th time I've said this now. I will not repeat it anymore.
 
Brent_Justice said:
It is not arbitrary. Nor does it favor one or the other. The ASUS card is not being held back in its overclock or its performance. This is about the 4th time I've said this now. I will not repeat it anymore.
Click to expand...

He won't be satisfied by anything that doesn't show the 290X coming out on top, it doesn't really matter what anyone says.
 
Memory controller is not the issue.

Many Hynix chips based cards like Tri X for example are hitting 6000 + as average, few of them go to 7000....

Asus use Elpida crap, they dropped the ball this generation sadly.
 
psyside said:
Memory controller is not the issue.

Many Hynix chips based cards like Tri X for example are hitting 6000 + as average, few of them go to 7000....

Asus use Elpida crap, they dropped the ball this generation sadly.
Click to expand...

eh my reference Sapphire 290X has Hynix memory and it tops out at 1450 MHz (5800 MHz). if i push it any more i get artifacts in some games. I keep it at 1400MHz for good measure. not all Hynix equipped cards are beasts and not all Elpida equipped cards are awful.
 
Run a custom ROM for the 780 TI and you get an even better overclock out of it.
 
Nice review guys.

The high temps on the ati cards, a nm shrink would help that wouldn't it? Maybe if they can do 10nm and chop the watts in half (gross generalization I know) it might have some OC headroom.
 
Great review as always! Thanks for putting in the time.

Green means go, Red means...? I guess we'll have to wait for other non-reference 290X models to appear and see what creative implementations are instituted. If anything can be done to combat the sheer amount of heat being generated.

If the 780ti is still claiming wins against custom 290X offerings, then I am really that much more anxious to see Maxwell perform.
 
Nenu said:
He predicted that the maximum overclock will be less than on other sites because [H] have spent a long time using the card and have made sure it is rock stable all the time.
Whereas most reviews will do a comparatively quick overclock test.
Click to expand...
This is true. Those other "reviewers" rarely test those quick oc's for stability as well.
FarleyM said:
$699 at newegg... jesus christ. Will newegg include a cuddle after ass raping you with the 150 mark up?
Click to expand...
Haha.
 
psyside said:
Memory controller is not the issue.

Many Hynix chips based cards like Tri X for example are hitting 6000 + as average, few of them go to 7000....

Asus use Elpida crap, they dropped the ball this generation sadly.
Click to expand...

"Many" and "few" are the key words there. The typical Hawaii memory OC among end users seems to be around 1400 which is still lower than the 1500MHz that the Elpida chips are rated for. How is that not a memory controller limitation? Are you saying that most of the Elpida chips and even some of the Hynix chips are not able to hit their rated speeds?
 
I don't see why so many people are concerned about memory speed on the 290X. It has 320GB/sec stock.

Using 3DMark Firestrike Extreme as an example I only gained ~120 points (9005 to 9120) bumping RAM from 1250 to 1400. I gained 600 points going from 1000 to 1100 core (8400 to 9000).

The memory isn't really a limitation.
 
Brent_Justice said:
Tried it. Apparently you did not read page 3.
I also tried 100% fan at 1.35v. End result, it did not.
Click to expand...

I know that the amount of overclock can vary but you seem to have a very bad sample there.

I have a Gigabyte 290X oc windforce (no not a fan boy just upgraded from my 2x gtx 460s which were great)

I can hit 1115mhz with a voltage increase of just +25 milivolts the core runs at about 1.170volts not the 1.35volts you are using...no wonder your thermals are through the roof.(in fact that vcore was good upto a clock of 1170mhz....so if I'd bothered to mess further with such a low overclock then I'm certain I could have cut that voltage down somewhat)

At those settings with 100% fan temps are 50c not the 90c you have...something is very wrong there.

I've got to 1210Mhz using the maxvoltage offset of +100mv which during use equates to a vcore of between 1.22-1.25volts.....temps with 100% fan are 61-65c so again your thermals and vcore are way not standard.

Having said that I'm at the limit now....I've tried reducing the % of fan speed but if temps go above 65c its not stable at that clock speed...so I may have to rein that in a little....memory clocks appear good at 1450 but haven't tinkered too much there yet.

I noted some anomalies whilst tinkering....I noticed that if you don't disable overdrive in CCC then power tune keeps dropping the voltage causing instability(even when using a separate overclocking tool).....may be why you felt the need to zap upto 1.35volts?

I downloaded the latest beta of MSI afterburner and selected "force constant voltage" on in the settings...the only variance of vcore then was due to v drop due to load.....seriously that card you tested is naf or your methodology isn't right for Hawaii.
 
Last edited:
The article notes the stable overclock achieved after long term gaming, for weeks. It is the most in-depth and tested overclocking I've ever done to make sure the overclock is stable for longevity, on both cards.
 
technogiant said:
I know that the amount of overclock can very but you seem to have a very bad sample there.

I have a Gigabyte 290X oc windforce (no not a fan boy just upgraded from my 2x gtx 460s which were great)

I can hit 1115mhz with a voltage increase of just +25 milivolts the core runs at about 1.170volts not the 1.35volts you are using...no wonder your thermals are through the roof.(in fact that vcore was good upto a clock of 1170mhz....so if I'd bothered to mess further with such a low overclock then I'm certain I could have cut that voltage down somewhat)

At those settings with 100% fan temps are 50c not the 90c you have...something is very wrong there.

I've got to 1210Mhz using the maxvoltage offset of +100mv which during use equates to a vcore of between 1.22-1.25volts.....temps with 100% fan are 61-65c so again your thermals and vcore are way not standard.

Having said that I'm at the limit now....I've tried reducing the % of fan speed but if temps go above 65c its not stable at that clock speed...so I may have to rein that in a little....memory clocks appear good at 1450 but haven't tinkered too much there yet.

I noted some anomalies whilst tinkering....I noticed that if you don't disable overdrive in CCC then power tune keeps dropping the voltage causing instability(even when using a separate overclocking tool).....may be why you felt the need to zap upto 1.35volts?

I downloaded the latest beta of MSI afterburner and selected "force constant voltage" on in the settings...the only variance of vcore then was due to v drop due to load.....seriously that card you tested is naf or your methodology isn't right for Hawaii.
Click to expand...

Technogiant, your posts make me laugh. Here you are talking about "unfair" methodology but you have missed the glaring flaw in your argument; if you wanted to worry about fan speeds and make this "fair" we'd need to disable one of the fans on that custom 290x. I'm sure your fanboy heart would break seeing those test results, so why not just leave well enough alone? AMD got their asses handed to them. Maybe they'll have an answer next generation.

Move on.
 
These cards vary a lot. Bubbaboeyhtj has his 290x under water and he reached similar results as hardocp.

8-pac, who's a world class overclocker got the 290x to 1200mhz with 1.4 volts using a 4 slot air cooler.

Considering the clocks of the gtx 780 ti are also on the low side, it could be the testing methodology is particularly strenuous. I think going by forums results, both are capable of about 75mhz more. But how stable are these cards at such clocks? Bench stable isn't the same as game stable.

The testing methodology favored the 290x by far. Using a non-factory cooled 290x is giving Hawaii huge favors. Without it, the overclocks could be less and power consumption would be way way up, with the chip leaking excessive heat.

Hawaii is a good GPU no question and a great value, but its simply a less elegant solution than the gtx 780 ti. It loses in this overclock scenario by 11% on average. This isn't that big a deal, the bigger deal is that it does so consuming 100watts less power. This to me show that Hawaii in its original form was probably something like a 6970 in its earlier form. Something that was meant to be 10-15 percent slower than Titan. What AMD did was implement boost in a manner where it could compete with Nvidia top end products at the expense of efficiency and headroom. Hardocp's overclocking results show just how much pinned to the limit AMD cards are set at and how relaxed gk110 is.
 
Last edited:
tajoh111 said:
These cards vary a lot. Bubbaboeyhtj has his 290x under water and he reached similar results as hardocp.

8-pac, who's a world class overclocker got the 290x to 1200mhz with 1.4 volts using a 4 slot air cooler.
Click to expand...

I think you are going to see a lot more of this, Hawaii is going to be all over the place in terms of overclocking.
 
Hi Brent. I remeber some time back regarding the ArmA 3 benchmarks, that it would be included in future gpu tests being one of the most taxing games out there. I don't know if this has been answered before, but will ArmA make a comeback in future reviews?

Thanks
Good article as always btw
 
DejaWiz said:
Great review as always! Thanks for putting in the time.

Green means go, Red means...? I guess we'll have to wait for other non-reference 290X models to appear and see what creative implementations are instituted. If anything can be done to combat the sheer amount of heat being generated.
Click to expand...

You don't need to wait, Tri X is a total beast.
 
specops said:
"Many" and "few" are the key words there. The typical Hawaii memory OC among end users seems to be around 1400 which is still lower than the 1500MHz that the Elpida chips are rated for. How is that not a memory controller limitation? Are you saying that most of the Elpida chips and even some of the Hynix chips are not able to hit their rated speeds?
Click to expand...

Easy, the memory could be undervolted. Like i said, 6/5 Tri X cards for now, hit like 1600+ they all got Hynix. Many other referent 290/290X hit at least 1500 as average with Hynix, with Elpida average is 1400.
 
Daroller said:
Technogiant, your posts make me laugh. Here you are talking about "unfair" methodology but you have missed the glaring flaw in your argument; if you wanted to worry about fan speeds and make this "fair" we'd need to disable one of the fans on that custom 290x. I'm sure your fanboy heart would break seeing those test results, so why not just leave well enough alone? AMD got their asses handed to them. Maybe they'll have an answer next generation.

Move on.
Click to expand...

Hey laugh all you want...glad to provide the entertainment.
I just find it hard to understand how or why it appears that the Hawaii chip itself can show such large variation in overclocking ability....I mean a 20% increase in voltage for a 6% increase in out of the box settings as has occurred here is ridiculous...we expect some variation in overclock capabilities..but not this much.

Strikes me the reasons are more likely to be due to poor or restricted cooling solutions or incorrect overclocking procedures.

While I'm reticent to suggest that a benchmarking and overclocking guru like Brent may have overclocked wrongly the Hawaii chip and power tune is relatively new and everyone is learning.

I did note that unless you disabled overdrive in ccc and forced a constant voltage that powertune would vary the vcore and so cause instability even at a moderate overclock....and I offer this as a possible reason that people are getting varying results?....they may be wildly increasing vcore to overcome this instability rather than disabling this and forcing a constant voltage.

So anyway....laugh all you like I've got my 290x at 1210Mhz on air with a dual slot cooler at vcore 1.25vcore and temps approx 61c...so forgive me for calling into question the results here.
 
Last edited:
You must log in or register to reply here.
Back
Top